Question

A retail company is building a Copilot Studio agent that will autonomously approve or deny customer refund requests up to $500 and communicate the decision directly to customers. The architecture team is following Microsoft's Responsible AI Standard. During which phase should the team conduct the formal Responsible AI Impact Assessment for this agent, and what is its primary purpose?

Accepted Answer

During the initial design/planning phase, to identify intended uses, stakeholders, and potential harms so mitigations can be built into the solution before development

Answer

After the agent passes all end-to-end functional test cases, to confirm the accuracy metrics meet the acceptance threshold before production release

Answer

During post-deployment monitoring, to review telemetry and abandonment rates so harms can be documented after real customer interactions occur

Answer

Only if a compliance audit is triggered, to retroactively document how the agent's decisions were made for regulators